cosmicstack-labs/mercury-agent
Mercury Agent
Mercury Agent is a personal AI agent that runs as a long-lived background process on your own machine. It restarts after crashes and at login, keeps working on tasks after the terminal is closed, and serves every connected chat channel from one shared local memory. The same agent can be reached from a terminal, Telegram, Discord, Slack, Signal and a local web dashboard, with access rules such as pairing codes and admin roles applied per channel.
Every file edit, shell command, network fetch and git action is checked against folder scopes, a fixed blocklist and per-session approval modes before it runs. Language model requests go to a default provider first and fall back down a ranked list of hosted, local or OpenAI-compatible endpoints when one fails. A daily token budget is counted on every turn, and as usage approaches the cap the agent trims context or sends routine work to cheaper models. The agent's character and behavior are defined in plain Markdown files you can edit.
Beyond chat, it covers coding inside a repository, where it plans changes with a live checklist and reports success only after a build, test or typecheck passes. It can also create scheduled specialist bots with their own personas, models, memory, permissions and budgets. Work can be tracked on Kanban boards that run cards in dependency order, with per-card token tracking and recurring or one-time jobs that report back to the originating channel. Facts from conversations are stored in a local database and recalled before each answer. Community skills that follow the Agent Skills specification can be searched, reviewed and installed to extend what the agent can do.
